Click to edit Master text styles
Second level
Third level
Fourth level
Fifth level
‹date/time›
‹footer›
‹#›
------=_NextPart_01C67E6B.922979F0
Content-Location: file:///C:/9A585236/KirkpatrickCO306_files/master03.xml
Content-Transfer-Encoding: quoted-printable
Content-Type: text/xml; charset="utf-8"
‹header›
‹date/time›=
=
‹footer›
‹#›
=
------=_NextPart_01C67E6B.922979F0
Content-Location: file:///C:/9A585236/KirkpatrickCO306_files/pres.xml
Content-Transfer-Encoding: quoted-printable
Content-Type: text/xml; charset="utf-8"
![](3D"master03_image003.gif")
![](3D"master03_image004.gif")
The rol=
e of
temporal generalization in temporal <=
/span>discrimination tasks
Kimberly Kirkpatrick
and
Rona Russell
Peak Shift<=
span
style=3D'mso-special-format:lastCR;display:none'>
•<=
/span>Church and Gibbon (1982) trained rats=
on a
temporal gener=
alization
task and found an orderly symmetrical generalization gradient.
•<=
/span>Does temporal generalization obey the=
same
rules as gener=
alization
in other dimensions?
•<=
/span>Hanson (1959) – peak of the
generalization gradient shifts away from the Standard (S+) value when intr=
a-=
dimensional discrimination training is
administered.
•<=
/span>Direction of shift is usually away fr=
om the
Non-stand=
ard
(S-).
•
Peak Shift<=
span
style=3D'mso-special-format:lastCR;display:none'>
![](3D"slide0003_image006.gif")
Spetch, M. L. & Cheng, K. (1998). A s=
tep
function in pigeons’ temporal generalization in the peak shift task.
Animal Learning & Behavior, 26, 103-118.
Peak Shift and Temporal <=
span
lang=3DEN-GB style=3D'font-size:91%'>Generalization
•T=
he
classic peak shift effect doesn’t appear to occur in temporal generalization gradients.
<=
span
class=3DBB style=3D'position:absolute;left:-4.43%'>•Why?
•Spetch
& Cheng (1998) proposed that their pigeons
had encoded the short and long samples
categorically.Experiment 1
=
226;Four groups of rats; two-lever discrimination <=
span
style=3D'position:absolute;top:31.5%;left:9.73%;width:84.45%;height:6.75%'=
>task
=
226;All groups received a 20-s Standard duration
&#=
8226;Non-standard durations were different
&=
#8211;Control – no non-standard duration
&=
#8211;NS 10 – 10-s non-standard
&=
#8211;NS 40 – 40-s non-standard
&=
#8211;NS 10 & 40 – 10- and 40-s non-standards
![](3D"slide0006_image007.gif")
Peak shift
Acquisition=
![](3D"slide0007_image009.gif")
Generalization Gra=
dients
•Con=
trol
group produced a broad, asymmetri=
c =
gradient with a peak at 21.6 s&=
#13;
•Gro=
up NS
10 produced a sharp gradient w=
ith a peak at 32.0 s
•Gro=
up NS
40 produced a broader gradient=
with a peak at 11.6 s
![](3D"slide0008_image011.gif")
Experiment 1 Summa=
ry
•I=
nterval
duration was directly related to discrim=
ination
accuracy.
•Temporal peak shift effects d=
id
occur under the present
conditions.
•NS 10 and NS 40 produced ord=
erly generalization gradients.
•T=
he
results inconsistent with categorical encoding
the samples.
Experiment 2
•Examined failure of NS 10 & 40 to learn the task
•Used more widely spaced intervals
–(10 s, 40 s, or 100 s)
•Further
assessments of the form of temporal generalization
•Three groups of rats
–SM/L (Group 1)
–SL/M (Group 2; same as NS 10 & 40)
–S/M/L (Group 3)
![](3D"slide0011_image013.gif")
10
s
40
s
100
s
40/100
> 10/40 > 10/100
=
Group SM/L (1)=
: 10/40 <=
span
lang=3DEN-GB style=3D'font-family:"Times New Roman"'>confusions will not r=
esult in
an=
error;
P(error) =3D P(40/100)
+ P(10/100)
Group
SL/M (2):
10/100 confusions
will not result in an
error; P(error) =3D P(10/40) + P(40/100)
=
Group S/M/L (3): P(error) =3D <=
span
lang=3DEN-GB style=3D'font-family:"Times New Roman"'>P(10/40) + P(40/100) =
+ <=
span
lang=3DEN-GB style=3D'font-family:"Times New Roman"'>P(10/100)
=
A Generalization A=
ccount
![](3D"slide0011_image014.gif")
![](3D"slide0011_image015.gif")
![](3D"slide0011_image016.gif")
Acquisition=
![](3D"slide0013_image018.gif")
SM/L (1) > SL/M (2) >=3D S/M/L (3)
Confusion Errors
•Pat=
tern
of confusions is perfectly consiste=
nt
with generalization gradients
=
•10/=
100
confusions are rare, so this may
explain the lack of difference between groups SL/M (2) and=
S/M/L
(3)
•Ther=
efore,
the generalization account =
is favored
![](3D"slide0014_image020.gif")
Testing
![](3D"slide0015_image022.gif")
![](3D"slide0015_image024.gif")
Gradient Summation=
•The=
sum
of the 10 s and 40 s gradients for=
S/M/L
(3) corresponds closely with the 10 s & 40 s gradien=
t in
SM/L (1).
•The=
100-s
gradients also correspond closely=
.
•This
suggests that the samples were treat=
ed independently.
Summary
•<=
/span>Experiment 1
–<=
/span>A standard peak shift effect was obse=
rved
–<=
/span>Orderly effects of interval duration =
on
discrimination accuracy
and spread of the gradients
•<=
/span>Experiment 2
–<=
/span>Wider interval spacing improved perfo=
rmance
on SL/M (2)
–<=
/span>The accuracy scores corresponded best=
with a generalization account
–<=
/span>Discriminal distance scores were cons=
istent
with scalar generalization
gradients as were the test functions in S/M/L
–<=
/span>Generalization testing indicated that=
10-,
40- and 100-s samples
were responded to independently
Summary
•O=
verall,
there was no indication of categorical encoding
in these tasks
•On the other hand, a simple
generalization account with scal=
ar
variance was consistent with a=
ll of
the findings
•This suggests that temporal
generalization may mimic generalizat=
ion in
other stimulus dimensions=
‹header›
‹date/time›=
Click to edit Master text styles
Second level
Third level
Fourth level
Fifth level
‹footer›
‹#›
After 40 sessi=
ons of
training; NS 40 required additional training to meet criterion;
all rats in this group eventually did meet criterion. Discriminability is =
a direct function of interval duration.
NS 40 not real=
ly a step
function; systematic decrease in S responses between S
and NS; tail at 80 s
Well-known effects of t=
est
range on shape of generalization gradients; rest=
ricted
range is more likely to yield a step function.
After 100 sessions of training, 1 > 2=3D3; 2 an=
d 3 are
performing poorly.
Why are 2 and 3 not different. Least problematic for a
generalization account because 2 vs.3 difference=
is
relying on the 10/100 errors in Group 3 to retard performance.
Group SM/L (1)=
tested
5/6 rats; Group SL/M (2) tested 0/6 rats; Group S/M/L (3)
tested 3/6 rats.
' + ENDSHOW_MESG + '
![](3D"notes_flag.gif")
No=
tes![](3D"buttons.gif")
![](3D"slide0016_image026.gif"){kind=link}